Creating Your First Project

A project serves as a container for managing a collection of images, annotations, and models. It provides a structured environment to streamline the process of detection, annotation, and model training.

graph TD
    A[Create a New Project] --> B[Add Media]
    B --> C[Run Detections]
    C --> D[Annotate Media]
    D --> E[Train Models]
    E --> B
    D --> F[Export Annotations]

Configuring Your Project

Configuring a project involves defining the following settings. You will be prompted to do so on creation of new projects.

Setting	Description
Load model	Select what model will be used for detections.
Specify detections	Select which objects the model should write detections for.
Model confidence	How certain the model should be in the object in order to write the detection.
Select video slice interval	In the case of videos, how far apart should the frames be captured?
Automatically filter dead video	Select whether the video should save frames that do not include a written detection.

Info

These settings may be changed at any time after initial configuration by navigating to the "settings" tab while in a project.

Tip

Remember to press "Apply" when you are done!

Adding media

After configuration, you can begin to add media. Currently, we support the following formats:

• .MP4
• .MOV
• .PNG
• .JPG

Info

Media will be screened on upload, before being loaded into the project. Detection may be re-run on a per-frame basis with a custom confidence via the "detect" button, in the event that experimenting with confidence is desired.

Warning

Running "detect" on a single frame will overwrite existing detections.

Annotating media

Annotations represent instances of a particular object within some image. These may be created by either users or, eventually by a model -- once the user has a trained model, that is.

Since we may expect some degree of errors in model detections, users may wish to correct the result of a model's work. That is to say: they may wish to add, remove or edit annotations. This functionality is available within currently active projects, either by selecting the + New Annotation option or using the shortcut: N. Annotation creation is by drag-and-drop, and the object category ("class") may be selected either by rotating through options via the Space key, or manually selecting the type of object in the top-left dropdown Class selection.

If you are unhappy with a particular annotation, you may delete it with the Del or Backspace keys.

Warning

Additional object types may not be added after project creation.

Editing Media

While in a project, the Edit Media button may be selected to bring up a dialogue showing all media in the project. From here, users may quickly select a variety of images for removal from the project, or clear the project entirely.

Model Training

While training is possible on-CPU, it is strongly recommended to use a CUDA-compatible GPU to ensure time-efficient training.

Additional dependencies are installed during first training session.

Warning

An internet connection is required for first training session in order to download the required depencies for training.

Users must provide a location for the model to be saved and also have the option to configure advances settings such as base model for an initial model to be used as a starting point, as well as several more parameters for fine-tuning your training session.

Exporting Data

Project data can be exported in several formats, from within the File menu. The options are as follows:

1. Export annotation data exports annotations in raw format (CSV, YAML, COCO or JSON) for further analysis.

2. Additionally, media may be exported one-by-one with the annotations drawn-on for later human reference. Selecting this option will open a file selection dialogue where users may select the desired images to export.

Info

Both aforementioned exporting techniques allow for filtering by class type - allowing for the minimum amount of data to be saved - as well as exporting without any filter applied.